This dataset is designed for CausalLM instruct fine-tuning, optimized specifically for 2B and other small language models.
It focuses on clean reasoning supervision without exposing long chain-of-thought, making it suitable for stable and efficient training.
Each sample follows this unified schema:
{
"instruction": "Solve the problem.",
"input": "Problem statement or question.",
"output": "Short answer or concise… See the full description on the dataset page:
https://huggingface.co/datasets/boopathiraj/Instruct_Dataset.
